跪拜 Guibai
← Back to the summary

AI Subscriptions Are the New Unreimbursed Work Expense

Before, the company gave you a computer.

Later, the company gave you cloud servers.

And today, a new question has emerged:

If the company doesn't provide a large language model, will you buy one yourself?

If the answer is yes.

Then congratulations.

The era of paying to work has truly arrived.

Image description

As ChatGPT, Claude, and Gemini become productivity tools, something is quietly happening

More and more programmers, product managers, operations staff, designers, testers, and even HR are silently paying out of their own pockets.

ChatGPT Plus, Claude Pro, Gemini Advanced, Cursor Pro, GitHub Copilot…

Tens of dollars or over a hundred yuan per month doesn't seem expensive.

But have you ever considered one question?

Who should be paying for these?

In theory, they are no longer entertainment expenses but production tools.

Just like Photoshop, IntelliJ IDEA, Office, or Figma.

Yet the reality is:

Many people are paying out of their own pockets to buy productivity for their companies.

This is a brand new era.

A programmer might subscribe to 5 AI tools per month

A developer's gear used to be simple:

And today?

Many people's toolchains now look like this:

Claude Pro
↓
Handles architecture design

Cursor Pro
↓
Handles writing code

ChatGPT Plus
↓
Handles problem analysis

GitHub Copilot
↓
Handles code completion

Perplexity
↓
Handles research

If you add:

A developer could spend thousands of yuan a year.

Here's the problem.

Most of these tools are not for personal life, but for work.

Image description

Even more absurd: Those who don't use them see their competitiveness decline

Before, working overtime versus not working overtime was just a difference in working hours.

Now.

Using AI and not using AI have become two different levels of production efficiency.

Take the simplest example.

Writing a CRUD feature before:

Half a day.

Today:

Claude + Cursor

10 minutes.

Debugging a bug before:

Google for half an hour
Stack Overflow for half an hour

Today:

Throw the logs at Claude.

3 minutes.

Writing a technical proposal before:

Half a day.

Today:

ChatGPT

20 minutes.

So, a strange phenomenon has appeared.

Not using AI doesn't mean you lack ability.

But it means you are getting slower and slower.

And in the workplace, the competition is often not about ability, but about the ability to create value per unit of time.

AI is becoming the new "office software"

Twenty years ago.

If you didn't know Office, it was hard to find a job.

Ten years ago.

If you didn't know Git, it was hard to find a job.

Today.

Not knowing how to use large language models is starting to affect work efficiency.

The real change isn't:

Will AI replace programmers?

But rather:

Programmers who can use AI will replace those who can't.

This is no longer just a slogan.

It is becoming reality.

Why haven't companies fully reimbursed AI expenses?

There are actually a few practical reasons for this.

First, the budget system hasn't caught up

Office has a procurement budget.

Adobe has a procurement budget.

JetBrains has a procurement budget.

But many enterprises:

Have no AI software procurement budget.

Because this category never existed before.

Many companies don't even know:

What kind of expense is Claude?

IT?

Software?

Office?

R&D?

So.

The simplest method is:

Employees buy it themselves.

Second, many companies haven't realized the ROI

What the boss sees is:

20 dollars per person per month.

But doesn't see:

Two hours saved per day.

Assume:

A developer's monthly salary is 20,000 yuan.

Saving 2 hours a day.

Saved in a year:

2 × 22 × 12

≈ 528 hours

Equivalent to:

Three months of working time.

And the cost of AI:

Less than two thousand yuan a year.

Many enterprises haven't truly calculated this account.

Third, security and compliance concerns

Many companies worry about:

Therefore, they prefer to delay deployment rather than open access rashly.

This is also why more and more enterprises are starting to choose:

In the future, what companies actually purchase might not be ChatGPT, but their own AI infrastructure.

What's really changing isn't AI, but the cost structure of work

Before, the company provided:

In the future, the company will also need to provide:

These will gradually become infrastructure, just like the network, water, and electricity.

Employees should not have to pay out of their own pockets for key production tools just to be efficient at work.

Even more worth pondering: AI is becoming a new "workplace ticket"

In the past, job requirements might have been:

Proficient in Java
Proficient in Linux
Proficient in MySQL

In the future, they are likely to become:

Proficient in using Claude
Proficient in using Cursor
Possess AI Coding capabilities
Familiar with Agent workflows
Able to build MCP toolchains

Not being able to use AI, just like not knowing Git or Docker back in the day, is gradually becoming a shortcoming.

AI is no longer just a tool, but a fundamental professional skill.

Final thoughts

"Paying to work" sounds like a joke.

But it reflects a reality that is unfolding:

AI has gradually evolved from a "nice-to-have" efficiency tool into the infrastructure of modern work.

When more and more people subscribe to ChatGPT, Claude, Cursor, and Copilot at their own expense just to get their work done, it not only shows that AI's value has been recognized, but also that the enterprise software procurement model is undergoing a new transformation.

In the future, we might no longer discuss:

"Do you use AI?"

But rather:

"Does your company provide you with AI?"

Because what truly efficient teams compete on is not just excellent talent, but whether that excellent talent has equally excellent production tools.

When computers, IDEs, and cloud servers are all provided by the company, large language models will eventually become a standard corporate provision.

And before that happens, the era of paying to work may have truly already begun.

Comments

Top 1 from juejin.cn, machine-translated. The original thread is authoritative.

腹黑的猫

Now it's bring your own computer, bring your own token.